Playing with replicated murder

[Ken Murchison]

Christoph Moench-Tegeder wrote:

> Hi,
> my employer wants to spend some money and buy a cluster filesystem
> for our IMAP servers (not that we currently need it, with only about
> five thousand active users on the largest server).
> The powers that be want to have "high availability" and "load balancing"
> (what load?). This sounds like a replicated murder to me.
> We already do have a SAN (fiberchannel and some spare gigabytes on an EMC)
> and a cluster filesystem (currently we are evaluating Polyserve, has
> anyone had some experience with psfs?).
> I am a little confused over the location of configdirectory, some of its
> contents (like the DB environment, the socket und the proc directorys)
> should be kept per node (I believe), while others (as quota files)
> should be shared between nodes. What about the replicated mailboxes.db?
> What am I missing?

I'm assuming that your talking about the current code in 2.3 which uses 
MUPDATE to replicate mailboxes.db across multiple backends sharing the 
same spool.

mailboxes.db is local to each machine, as are deliver.db and tls_sessions.db

The user's seen state and subscriptions, as well as quotaroots can be 
shared on the SAN.

So basically /var/imap is local, but /var/imap/user, /var/imap/quota, 
/var/spool/imap are on the SAN.

Note that the university that I wrote the code for has abandoned it 
because of problems with Sun's SAN filesystem.

-- 
Kenneth Murchison     Oceana Matrix Ltd.
Software Engineer     21 Princeton Place
716-662-8973 x26      Orchard Park, NY 14127
--PGP Public Key--    http://www.oceana.com/~ken/ksm.pgp
---
Cyrus Home Page: http://asg.web.cmu.edu/cyrus
Cyrus Wiki/FAQ: http://cyruswiki.andrew.cmu.edu
List Archives/Info: http://asg.web.cmu.edu/cyrus/mailing-list.html